Geoinformation, monitoring and assessment Environment

Posted March 1996

A Methodology for Monitoring Wetlands for Fisheries Using NOAA AVHRR LAC Thermal Data

Carlo Travaglia
Remote Sensing Officer
Environment and Natural Resources Service (SDRN)
FAO Research, Extension and Training Division
James McDaid Kapetsky
FAO Fishery Resources and Environment Division
Gaia Righini
FAO Remote Sensing Consultant

(extracted from RSC Series 68, "Monitoring Wetlands for Fisheries By NOAA AVHRR LAC Thermal Data" FAO, 1995.)


Wetlands are economically important and sensitive ecosystems usually covering extensive areas, but subject to large variations - both seasonally and inter-annually - according to the climatic conditions of the region involved. The fishery potential of wetlands is directly related to the duration and extent of the floods, as well as to the efficiency of fishing. Consequently, fishery potential varies from season to season and from year to year. Frequent baseline data on wetlands are essential for assessing fishery potential and for a host of hydrologic and environmental applications.

Remote sensing is the only tool which can be used to provide those data at frequent intervals and economically. Thus an operative methodology based on the thermal inertia difference between dry and wet lands has been developed from NOAA (National Oceanic and Atmospheric Administration) satellite thermal data. Wetlands in Zambia and Sudan were monitored through a selective time series of data. Results are in line with available ground estimations and show the great time variability of those areas.

NOAA satellite data

Satellite remote sensing is unique for assessing large wetland areas and monitoring their surface changes at periodic intervals, avoiding expensive, difficult and often useless ground surveys.

NOAA satellites are meteorological-environmental satellites placed in a near-polar, sun-synchronous orbit. The AVHRR (Advanced Very High Resolution Radiometer) they carry on board has a 10 bit radiometric resolution, five spectral channels (Table 1) and a spatial resolution of 1.1 km (LAC- Local Area Coverage), which implies images with a regional synoptic view, suitable for mapping at the scale 1:1,000,000. The NOAA systems also have a high rate repetition coverage, with daily passes in early morning and early afternoon.

Table 1 - AVHRR Spectral Channels
Ch 10.58 - 0.68 micron visible
Ch 20.725 - 1.10 micron near infrared
Ch 3 3.55 - 3.93 micronlow thermal infrared
Ch 410.3 - 11.3 micronthermal infrared
Ch 511.5 - 12.5 micronthermal infrared

NOAA AVHRR data provide the advantage of global coverage at high temporal resolution: they are very useful in monitoring natural resources or phenomena where frequent and large area mapping is required and fine detail is not. Furthermore, the coarse spatial resolution reduces the amount of data to be processed; the low cost of a single NOAA scene, compared with that of other satellite data, is certainly another advantage.

Other satellite systems traditionally used in natural resources assessment and monitoring such as SPOT and Landsat TM were considered not useful in this study. Actually, due to their high resolution (10 to 30 m) and the coverage of each scene, many images would have been necessary to cover the areas of interest, with corresponding high costs. Discrimination of the land/water interface would also have been difficult owing to the occurrence of vegetation on both land and at the water's surface.

Further, to apply the remote sensing methodology utilized in this study and described in the next paragraph, data in the thermal portion of the electromagnetic spectrum , recorded around noon, are necessary. As the thermal channel of Landsat TM (bd 6) is recorded around 9.30 hours local time, these data were not useful for the study.

Thermal inertia approach

In order to discriminate wetland areas and monitor their surface changes, the thermal inertia of bodies, as derived from remotely sensed data, has been utilized for the pilot study. Thermal inertia of bodies is a measure of their resistance to changing temperature under a time varying energy flux and therefore it is proportional to the time involved in absorbing and losing heat; it increases with an increase in material conductivity, capacity and density.

Materials with high thermal inertia have more uniform surface temperature throughout the day and night than materials of low thermal inertia. Water and aquatic vegetation have higher thermal inertia than that of dry land around and of non-aquatic vegetation as the former warm up less during the day; thus water and vegetation floating on it or living in it normally appear cooler than their surroundings on daytime thermal images.

Therefore, although a traditional approach in evaluating thermal inertia involves day and night coverage, it is possible to discriminate between surfaces with different thermal inertia using the thermal portion of the electromagnetic spectrum with coverage around noon: this is when the thermal contrast between dry and wet areas reaches the maximum as demonstrated by Mason et al. 1992. In the study of the Sudd swamp, using the thermal channel of the Meteosat they produced the graph reported in Fig. 3 which is self explanatory.

The thermal channel 4 of NOAA AVHRR LAC was found as the most suitable for thermal inertia studies.

Fig. 1 - DN values taken from Meteosat images showing thermal contrast variation around the diurnal cycle (Mason et al., 1992)

Data selection

The principal task was to select data recorded through the years according to the rainfall and hydrological cycles of the regions in order to monitor the seasonal and inter-annual variations of the inundated areas. In order to achieve this, NOAA AVHRR quick-looks were used to find cloud-free images, or those with as little clouds as possible during both the rainy and dry seasons. The data listed in Table 2 were thus acquired.

Table 2 - List of NOAA AVHRR data used
14 April 199213.0630 December 199112.30
3 June 199213.107 January 199212.37
19 June 199213.308 April 199212.53
7 July 199213.1328 September 199213.31
24 August 199213.3231 December 199213.04
7 June 199314.2130 March 199313.33
9 July 199314.4510 September 199313.46
30 July 199313.509 December 199313.53
22 August 199314.0119 January 199413.55
15 April 199414.3114 March 199414.39
29 April 199415.12
10 June 199415.03
26 June 199415.50
21 August 199415.08

Unfortunately not all the data were as useable as expected: in some cases, even if the wet region itself was not covered by clouds, they occupy a large part of the scene, completely dominating the spectral response of the satellite sensor while DN (Digital Number) values relating to the other features are limited to a very narrow range of the histograms. This problem particularly affects some images of Lake Bangweulu in Zambia, where, in most cases, it seems quite difficult to resolve ambiguities.

Data processing

Each scene covers the whole of Zambia or a large part of Sudan and Ethiopia, but not in the same way and often with a considerable spatial distortion. Thus the first task was thus to geo-reference all images to the Latitude/Longitude Projection using ONC maps (scale 1:1.000.000), in order to be able to identify areas of interest and to compare different data sets.

Difficulties were encountered in identifying ground control points on the images due to the low spatial resolution; thus for this purpose points located on country boundaries were utilized, overlaid during preprocessing on the row image by the supplier. The row data are, actually, difficult, if not impossible, to use. That is why, also due to the large area under consideration, sometimes images could not be perfectly superimposed; the error however is in the order of no more than a few pixels.

The whole data set was processed, using the same methodology, in order to develop the most suitable image for visual interpretation; attention was focused on thermal band 4 where the contrast between dry and wet areas is highest.

Considering the statistical values of the images, the best choice is to display them utilizing one standard deviation from the mean. After that a thresholding technique was carried out: all DN values of land vegetation and dry areas were saturated, leaving only open water surfaces or water covered by aquatic vegetation; this involved a trial and error approach based on the study of the image's brightness histograms. As a result, the wetter versus drier areas were emphasized, with the colour scheme convention used for meteorological satellite that is cooler areas appear lighter toned (such as wetlands) and warmer areas appear dark toned (bare soil).

Other enhancement techniques were then tested such as edge detection, convolution filtering and histogram equalization, but without any appreciable improvement. Also other bands were analysed in order to obtain as much information as possible from satellite data: in some cases thermal band 3 was useful to clarify ambiguities among different features.

The NDVI (Normalized Difference Vegetation Index) also was displayed by combining data from visible and near infra-red channels: in this way it was possible to emphasise vegetated areas (in white and light grey) from non vegetated ones whatever they are. NDVI and thermal inertia from channel 4 provide different but converging information for the discrimination of all features of interest such as open water, water covered by floating vegetation and bare soil.

Having identified the wetlands on the images, estimates of the surface areas inundated area were made. The results, listed in Tables 3, 4, 5 and 6, should not be considered as exact because the flood area measurement is necessarily achieved by contouring the inundated area on the screen, but are close to the actual surface areas. The variability in flooded surface area, both seasonally and inter-annually, is evident.

Table 3 - Barotse flood plain area variation
April, 14/19923600
June, 3/19922580
June, 19/19922500
August, 24/19922200
June, 7/19935200
July, 9/19935000
July, 30/19934500
August, 22/19934000
April, 29/19945000
June, 10/19944200
June, 26/19944200
August, 21/19943900

Table 4 - Kafue Flats, Itezhi Tezhi Reservoir flood plain area variation
April, 14/19921600
June, 3/19921500
June, 19/19921390
July, 7/19921090
August, 24/19921000
June, 7/19935100
July, 9/19934900
July, 30/19934900
August, 22/19934600
April, 15/19944100
April, 29/19944100
June, 10/19945200
June, 26/19944800
August, 21/19944500

Table 5 - Lake Bangweulu Swamp area variation
June, 3/19926 200
June, 19/19926 000
July, 7/19926 000
August, 24/19925 600
August, 22/19936 500

Table 6 - SUDD flood plain area variation
December, 30/9146000
January, 7/9248000
April, 8/9228000
September, 28/9236000
March, 30/9334000
September, 10/9331000
January, 19/9436000
March, 14/9429000


The results obtained in this pilot study have clearly demonstrated the possibility of monitoring wetland systems using NOAA AVHRR thermal data. The remote sensing methodology based on the difference in thermal inertia between dry and wet land has permitted the assessment of inundated areas mostly covered by aquatic vegetation. Seasonal and inter-annual variability of the wetland systems considered in this study are evident on the enhanced NOAA AVHRR images; further, the inundated area variation has been estimated and the results reported in km2 in the tables. The NDVI has provided complementary information, allowing for the discrimination of open water, aquatic vegetation and bare soil within the wetland systems.

Comparison of NOAA AVHRR interpretation results with available information for the study areas has been made: remotely sensed area estimates are well in the range of published data, if the seasonal and inter-annual variability of the wetland systems is considered.

Unfortunately, there is no easy way of assessing the absolute accuracy of the area estimates, as several difficulties prevent ground control surveys. To validate the methodology on the ground, we are planning to locate some points along the land/water interface in the Lake Bangweulu wetland system by GPS (Global Positioning System). A NOAA AVHRR image, acquired in the same period of the field verification, will be processed according to the methodology described and the land/water interface on the image and that resulting from GPS interpolation will be compared. This validation campaign is scheduled for early 1996.

Back to Top FAO Homepage